A drone-mounted camera recognized American Sign Language in early tests

The system used a flying camera and onboard computing to recognize gestures in less than two seconds on average.

Editorial illustration — not from the study.

SIGN-AIR combines a stabilised RGB camera mounted on an aerial drone with onboard computing. It extracts hand movements and uses an artificial intelligence system to analyze both the shape and timing of American Sign Language gestures in indoor and outdoor settings.

In tests using the ASL Alphabet and WLASL datasets, the system achieved 98.63% overall recognition accuracy, with a median response time of less than two seconds. The researchers describe the bidirectional translation system as a possible tool for accessibility, education, public services and emergency response, but say it still needs testing in real-world conditions.

What the drone recognized

The system recognized American Sign Language gestures with 98.63% overall accuracy on a combined dataset made from the ASL Alphabet and WLASL datasets. Its median inference latency—the time needed to produce a result—was less than two seconds. The evaluation also reported high precision and recall, although the abstract does not give their numerical values. SIGN-AIR includes a proposed bidirectional translation pipeline intended to support both sign-to-speech and speech-to-sign communication.

Why a flying camera matters

A drone-mounted camera could move beyond the limited view of a fixed camera and operate in places where communication infrastructure is unavailable. That flexibility may be useful in indoor or outdoor settings affected by occlusions, restricted camera coverage or changing conditions. The work points to possible uses in accessibility, education, public-service interactions and emergency response, while not yet showing that the system works reliably in those settings.

Evidence and open questions

The findings come from an experimental evaluation on combined sign-language datasets, rather than from comprehensive field tests with a drone operating in representative real-world conditions. The study demonstrates technical feasibility under the tested conditions, but the abstract says further validation is needed. It does not establish how performance would change with different lighting, backgrounds, camera motion, distances, signers or operational environments.

Fine-particle levels around cement plants were above the WHO annual guideline

A meta-analysis found that fine-particle concentrations around cement plants had a weighted average of 20.76 micrograms per cubic metre, above the World Health Organization’s annual guideline of 5 micrograms per cubic metre. The researchers cautioned that the studies differed substantially, including in how long they sampled the air.

7 hours ago·1 views

AI & Computing

AI recognizes writing on severely damaged South Indian palm leaves

An AI system designed for damaged South Indian palm-leaf manuscripts reported 97.4% word accuracy and 98.6% character accuracy on an archival corpus of 1,250 folios. It also reported 99.4% accuracy in identifying authentic palm-leaf manuscripts.

3 days ago·10 views

Abstract editorial illustration accompanying the article: Why “Birds have hollow bones” can teach broad facts despite different interpretations

AI & Computing

Why “Birds have hollow bones” can teach broad facts despite different interpretations

People can interpret the same general statement in different ways, yet generic language may still pass on broad knowledge. Across three experiments, a model in which such statements have an uncertain threshold best explained participants’ varied interpretations of new generic claims.

4 days ago·28 views